Synthesizing Ginsenoside Rh2 in Saccharomyces Cerevisiae Cell

Wang et al. Cell Discovery (2019) 5:5 Cell Discovery DOI 10.1038/s41421-018-0075-5 www.nature.com/celldisc

ARTICLE Open Access Synthesizing ginsenoside Rh2 in Saccharomyces cerevisiae cell factory at high-efﬁciency Pingping Wang1,WeiWei1,WeiYe2,3,XiaodongLi1,2, Wenfang Zhao1,ChengshuaiYang1,2, Chaojing Li1,2, Xing Yan1 and Zhihua Zhou1

Abstract Synthetic biology approach has been frequently applied to produce plant rare bioactive compounds in microbial cell factories by fermentation. However, to reach an ideal manufactural efficiency, it is necessary to optimize the microbial cell factories systemically by boosting sufficient carbon flux to the precursor synthesis and tuning the expression level and efficiency of key bioparts related to the synthetic pathway. We previously developed a yeast cell factory to produce ginsenoside Rh2 from glucose. However, the ginsenoside Rh2 yield was too low for commercialization due to the low supply of the ginsenoside aglycone protopanaxadiol (PPD) and poor performance of the key UDP- glycosyltransferase (UGT) (biopart UGTPg45) in the final step of the biosynthetic pathway. In the present study, we constructed a PPD-producing chassis via modular engineering of the mevalonic acid pathway and optimization of P450 expression levels. The new yeast chassis could produce 529.0 mg/L of PPD in shake flasks and 11.02 g/L in 10 L fed-batch fermentation. Based on this high PPD-producing chassis, we established a series of cell factories to produce ginsenoside Rh2, which we optimized by improving the C3–OH glycosylation efficiency. We increased the copy

1234567890():,; 1234567890():,; 1234567890():,; 1234567890():,; number of UGTPg45, and engineered its promoter to increase expression levels. In addition, we screened for more efﬁcient and compatible UGT bioparts from other plant species and mutants originating from the direct evolution of UGTPg45. Combining all engineered strategies, we built a yeast cell factory with the greatest ginsenoside Rh2 production reported to date, 179.3 mg/L in shake ﬂasks and 2.25 g/L in 10 L fed-batch fermentation. The results set up a successful example for improving yeast cell factories to produce plant rare natural products, especially the glycosylated ones.

Introduction microbial cell factories3,4, highlighting the potential for High-efﬁciency production of artemisinic acid via yeast metabolic engineering of microbial chassis to provide fermentation has become a milestone for the application reliable approaches for obtaining natural plant products of synthetic biology to the production of natural plant on a large scale, especially rare natural products. products1,2. Since then, a series of plant natural products Ginsenoside Rh2, derived from Panax species, is a pro- – have been successfully heterologously produced in mising candidate for cancer prevention and therapy5 7. However, the ginsenoside Rh2 accounts for less than 0.01% of dried Panax ginseng roots8. Commercially available gin- Correspondence: Xing Yan ([email protected]) or Zhihua Zhou senoside Rh2 is currently produced mainly via the ([email protected]) 1CAS-Key Laboratory of Synthetic Biology, CAS Center for Excellence in deglycosylation of major protopanaxadiol (PPD)-type gin- Molecular Plant Sciences, Institute of Plant Physiology and Ecology, Shanghai senosides isolated from Panax spp. using chemical or bio- – Institutes for Biological Sciences, Chinese Academy of Sciences, 300 Fenglin Rd, transformation approaches9 11.However,suchapproaches Shanghai 200032, China 2University of Chinese Academy of Sciences, Beijing 100049, China rely heavily on the cultivation of Panax plants and the Full list of author information is available at the end of the article.

© The Author(s) 2019 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a linktotheCreativeCommons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Wang et al. Cell Discovery (2019) 5:5 Page 2 of 14

successive cultivation of Panax plants is unsustainable, convert PPD into ginsenoside Rh2 increased by more than resulting in low productivity and vulnerability to disease 1800-fold. The yeast cell factory with this engineered outbreaks12. A synthetic biology strategy to artificially pro- glycosyltransferase could produce ~300 mg/L of ginse- duce ginsenoside Rh2 from glucose in a metabolically noside Rh2 in a 5L fed-batch fermentation system. engineered microbial cell factory is expected. However, there are still many disadvantages of the use of PPD is the common precursor of all dammarane-type UGT from bacteria or yeast, such as its poor substrate ginsenosides, including ginsenoside Rh2. A key enzyme to specificity. For example, the engineered UGT51 can also glycosylate PPD to produce ginsenoside Rh2 is required. catalyze the C3–OH glycosylation of many native yeast A Korean research group reported the isolation of two sterols (e.g., ergosterol), and thus produce diverse by- glycosyltransferase (UGT) genes involved in the bio- products and waste UGT enzyme activity17. Meanwhile, synthesis of the ginsenosides Rh2 and Rg313. Nearly Bs-YjiC exhibits poor site specificity to C3–OH, resulting concurrently, we also characterized two UGTs (UGTPg45 in lower ginsenoside Rh2 yield (0.2 g/L) compared with and UGTPg29) taking parts in the biosynthesis of ginse- the high F12 yield (3.98 g/L)16. nosides Rh2 and Rg3 independently14. In this previous Besides, optimization of the PPD-producing chassis may work, we constructed the strain D20RH18 for de novo also improve ginsenoside Rh2 yield. In the past several production of ginsenoside Rh2 from glucose based on the years, much progress has been made in engineering PPD-producing chassis strain ZW-PPD-B (chassis 1.0, microbes for high-level production of PPD. For example, with a PPD titer of 75.3 mg/L in shake flasks)14. However, by integrating the PPD biosynthesis pathway into a yeast the ginsenoside Rh2 production of D20RH18 was rela- chromosome and overexpressing the rate-limiting gene tively low (16.95 mg/L in shake flasks), and the yield of tHMG1 (truncated HMG-CoA reductase gene) and three total dammarane-type triterpenoids (dammarenediol II key precursor genes, the resulting PPD-producing strain [DM], PPD, and ginsenoside Rh2) was also low. Kinetic could produce 148.1 mg/L PPD in shake flasks, and the analysis of the biopart UGTPg45 to produce ginsenoside titer reached more than 1 g/L in fed-batch fermentation18. Rh2 revealed a low catalytic efficiency, only 1/2500 of that Recently, to overcome the poor coupling between P. of UGTPg29, which further catalyzes ginsenoside Rh2 ginseng cytochrome P450 and the Arabidopsis thaliana into Rg3. Therefore, we reasoned that the limited supply cytochrome P450 reductase ATR1, as well as the high DM of the precursor for dammarane-type triterpenoid synth- accumulation in PPD-producing yeast recombinants, the esis and poor performance of the key UGT biopart were fusion of PPDS and ATR1 encoding genes was introduced the two key factors hindering the high-efficiency pro- and the PPD production increased significantly, reaching duction of ginsenoside Rh2. 265.7 mg/L PPD in shake flasks and 4.25 g/L in 5 L fed- Since the first two reports of the ginsenoside Rh2 bio- batch fermentation, which is the highest PPD production synthetic pathway, several studies have attempted to ever reported19,20. optimize this glycosylation step by identifying UGT bio- In this study, we constructed a new PPD-producing parts from microbial sources or by engineering UGTs to chassis strain to improve the conversion of glucose into catalyze ginsenoside Rh2 formation. UGTs from bacteria PPD by overexpressing all mevalonic acid (MVA) pathway often show substrate flexibility and are potential candi- genes and optimizing the expression levels of cytochrome dates for ginsenoside Rh2 synthesis. UGT109A1 from P450 enzymes in yeast. In addition, we employed several Bacillus subtilis could catalyze the C3–OH and C12–OH strategies to optimize the cell factories to produce gin- glycosylation of DM, PPD, and protopanaxatriol (PPT)15. senoside Rh2 by improving UGTPg45 expression level Another UGT from B. subtilis, Bs-YjiC, which shares 94% and activity, including: (1) increasing UGTPg45 expres- identity with UGT109A1, could also catalyze the C3–OH sion by increasing its copy number and engineering its and C12–OH glycosylation of PPD to yield ginsenoside promoter and (2) increasing the in vivo activity of Rh2 and an unnatural ginsenoside F12 (3-O-β-D-gluco- UGTPg45 in yeast via protein engineering based on pyranosyl-12-O-β-D-glucopyranosyl-20(S)-proto- in vivo directed evolution and searching for novel UGTs panaxadiol)16. Coupling this enzyme with a UDP-glucose with higher C3–OH glycosylation efficiencies from other regeneration system powered by sucrose synthase, an plant species. in vitro one-pot ginsenoside synthesis platform was established to produce 0.2 g/L of ginsenoside Rh2 and Results 3.98 g/L of F12 using PPD as the substrate16. Another Design and construction of the PPD-producing chassis group built an efficient ginsenoside Rh2 biosynthetic cell 2.0 strain factory by repurposing an inherently promiscuous UDP- Ten enzymes are required to synthesize the triterpenoid glycosyltransferase UGT51 from Saccharomyces cerevi- precursor 2, 3-oxidosqualene from acetyl coenzyme A siae17. Based on a structure-guided semi-rational engi- (acetyl-CoA) in the MVA pathway. Then three hetero- neering approach, the catalytic efficiency of UGT51 to logous genes from P. ginseng, which encode three Wang et al. Cell Discovery (2019) 5:5 Page 3 of 14

Fig. 1 Schematic biosynthesis pathway of ginsenoside Rh2 in engineered yeast. Blue-colored genes form the upstream-module, red-colored genes form the downstream-module, and the green-colored gene is the UDP-glycosyltransferase biopart. IPP isopentenyl pyrophosphate, DMAPP dimethylallyl pyrophosphate enzymes that convert 2, 3-oxidosqualene into PPD, Meanwhile, the accumulated DM production in strain namely PgDDS, CYP716A47, and PgCPR1, are also ZW04BY reached to 195.9 mg/L (Fig. 2), and the total required for PPD biosynthesis. Therefore, we aimed to triterpenoid (PPD + DM) production of strain ZW04BY overexpress these 13 genes (Fig. 1 and Supplementary was 525.6 mg/L, a more than twofold increase compared Fig. S1). The 13 genes were divided into two groups. The with that of strain ZW03BY (240.6 mg/L). This indicated first group, designated as the upstream-module, included that introducing the upstream module boosted the seven upstream genes (ERG10, ERG13, ERG12, ERG8, metabolic flux toward the MVA pathway, although addi- ERG19, IDI, and tHMG1) that converted acetyl-CoA into tional efforts were required to convert the accumulated the isoprenoid building blocks isopentenyl phosphate DM into PPD. (IPP) and dimethylallyl phosphate (DMAPP). The second To further increase the conversion of DM into PPD, group, designated as the downstream-module, contained another copy of synPgPPDS was introduced into the the remaining six downstream genes (ERG1, ERG20, rDNA sites of ZW04BY to construct strain ZW04BY-RS, ERG9, synPgDDS, synPgPPDS, and synPgCPR1), which and PPD production reached 529.0 mg/L in shake flasks, converted IPP and DMAPP into the target product PPD. which was more than twofold that of strain ZW03BY. In The three heterologous genes PgDDS, CYP716A47, and addition, DM production decreased to 72.6 mg/L, repre- PgCPR1 were codon-optimized for expression in yeast senting only 12.1% of the total triterpenoid (PPD + DM) and renamed synPgDDS, synPgPPDS, and synPgCPR1, production (Fig. 2). respectively. Because HMG-CoA reductase catalyzes the These results suggest that the increasing expression of rate-limiting step in the MVA pathway, multi-copy key P450 genes could enhance the efficient conversion of overexpression of tHMG1 benefits the production of dif- DM into PPD. Moreover, the PPD titer of ZW04BY-RS ferent terpenoids1,21. Therefore, we added an additional was sixfold that of strain ZW-PPD-B and 1.99-fold that of copy of tHMG1 to the second group (Fig. 1). W3a, the highest PPD production strain constructed in After integrating the downstream-module into the delta the previous report19, representing the highest reported DNA site of the yeast strain BY4742, the resulting strain, production of PPD in a shake flask system. Therefore, ZW03BY, produced 2.7 mg/L DM and 237.9 mg/L PPD strain ZW04BY-RS might serve as a better chassis to from glucose in shake flasks (Fig. 2). This demonstrated produce ginsenoside Rh2. that our design should be feasible, because only a single- step integration was required to increase the PPD titer by Optimization of UGTPg45 expression in ZW04BY-RS more than threefold compared with that of ZW-PPD-B To assess the ginsenoside Rh2 biosynthesis efficiency of (i.e., chassis 1.0). In the next step, the upstream-module the new PPD chassis, UGTPg45 under the control of a was integrated into the Yprc-delta15 site of ZW03BY, strong constitutive promoter, TDH3, was introduced into yielding strain ZW04BY. Its PPD titer further increased to the single-copy X-4 site of ZW04BY-RS22,23. However, the 329.7 mg/L in shake flasks. However, the PPD titer of resulting strain ZWDRH2-1 produced only 35.7 mg/L of strain ZW04BY was only 39% greater than that of strain ginsenoside Rh2 in shake flasks (Fig. 3), and the Rh2 yield ZW03BY, and did not reach the expected target (Fig. 2). increased only 110% compared with that of strain Wang et al. Cell Discovery (2019) 5:5 Page 4 of 14

Fig. 2 Construction of the protopanaxadiol (PPD)-producing chassis 2.0 strain. a High-performance liquid chromatography (HPLC) analysis of dammarenediol II (DM) and PPD produced by ZW03BY, ZW04BY, and ZW04BY-RS. b Quantiﬁcation of DM and PPD produced by ZW03BY, ZW04BY, and ZW04BY-RS. The error bars indicate the SEMs of three biological replicates

D20RH18. In addition, the PPD yield of the newly built chromosome. The ginsenoside Rh2 yield of the resulting strain was 523.2 mg/L. This indicated that only a small strain ZWDRH2-2 increased to 52.7 mg/L (Fig. 3). Next, amount of PPD was converted into ginsenoside Rh2 in we introduced the strong artiﬁcial promoter UASTEF1- ZWDRH2-1, and its glycosylation ratio (i.e., Rh2/[PPD + UASCIT1-UASCLB2-TDH3 (hereafter UAS-TDH) con- Rh2]) was only 6.4%. Therefore, it was necessary to structed by Blazeck et al. to replace the TDH3 promoter24, increase the expression level or activity of yielding the strain ZWDRH2-3. As a result, the ginseno- the glycosyltransferase. side Rh2 yield further increased to 74.7 mg/L (Fig. 3). First, we attempted to increase UGTPg45 expression by Overall, ginsenoside Rh2 production was increased by introducing multi-copy of UGTPg45 into the delta DNA 109% by enhancing UGTPg45 expression level in the PPD sites, which has hundreds of copies in the yeast chassis ZW04BY-RS. Wang et al. Cell Discovery (2019) 5:5 Page 5 of 14

Fig. 3 Optimization of UGTPg45 expression in ZW04BY-RS. a HPLC analysis of DM, PPD, and ginsenoside Rh2 produced by ZWDRH2-1-3. b Quantiﬁcation of DM, PPD, and ginsenoside Rh2 produced by ZWDRH2-1-3. The error bars indicate the SEMs of three biological replicates

Screening for novel UGT bioparts with higher PPD catalysis UGTs has been applied in combinatorial biosynthesis and efficiencies to produce ginsenoside Rh2 glycol diversification of natural products27,28. Because of As reported in our previous work, the kcat/Km of the poor performance of UGTPg45 from P. ginseng,we − − UGTPg45 for PPD is 0.0018 μM 1 s 1, only 1/2500 that of searched for novel alternative UGT bioparts from other − − UGTPg29 for Rh2 (7.46 μM 1 s 1), and this poor catalytic plants to more efficiently convert PPD into ginsenoside activity has limited the biosynthesis of ginsenosides Rh2 Rh2. Three UGT genes, UGT73F3 from Medicago trun- and Rg3 in previously constructed strains14. As such, catula29, UGT73C10 from Barbarea vulgaris30, and UGTs with higher efficiencies are necessary to improve the UGTPn50 from Panax notoginseng, were characterized yield of ginsenoside Rh2. Many studies have demonstrated able to produce ginsenoside Rh2 in in vitro reactions by that some UGTs do not show strict substrate specificity, incubating the raw enzyme solution with PPD and UDP- only regio-specificity25,26. The substrate flexibility of these glucose (Supplementary Fig. S2). Wang et al. Cell Discovery (2019) 5:5 Page 6 of 14

To compare their ginsenoside Rh2 synthesis efficiencies, dimensional structures. Thus, an alternative strategy to the encoding genes UGT73F3, UGT73C10,andUGTPn50 screen for more efficient UGT was carried out by direct were introduced into the PPD-producing chassis ZW04BY- evolution of UGTPg45. Random UGTPg45 mutants in a RS at the single-copy X-4 site to construct strains library constructed with error-prone PCR were integrated ZWDRH2-4, ZWDRH2-5, and ZWDRH2-6, respectively. into the X-4 site of the PPD chassis ZW04BY-RS. Then, ZWDRH2-5 and ZWDRH2-6 produced 14.1 mg/L and colonies selected from culturing plates were incubated in 45.6 mg/L of ginsenoside Rh2, respectively (Fig. 4a); how- 96-well plates and subjected to fermentation. The ginse- ever, no Rh2 production was detected in strain ZWDRH2-4 noside Rh2 yield of each colony extracted with n-butanol owing to the low enzymatic activity of UGT73F3 toward in the 96-well plates was quantified by HPLC. After PPD even in the in vitro test (Supplementary Fig. S2). optimizing the HPLC running conditions, only 5 min was UGT73C10 can also catalyze the C3–OH glycosylation of required per sample, and a 96-well plate could be com- DM to produce 3-O-glucosyl-dammarenediol II (3-DMG) pleted within 8 h. in vitro, as shown in Supplementary Fig. S3, and the yield of Among the 16 plates of clones that were screened (> 3-DMG in strain ZWDRH2-5 was 9.3 mg/L. 1500 transformants), we acquired the transformant P8-E7 These results demonstrated that the introduction of (renamed strain ZWDRH2-8 hereafter), with a ginseno- UGTPn50 led to a higher ginsenoside Rh2 yield than side Rh2 yield of 60.5 mg/L, 1.7-fold that of strain UGTPg45, resulting in a 27.7% increase in production. ZWDRH2-1 with wild-type UGTPg45 (Fig. 4b). We per- Next,weperformedthecatalytickineticanalysisof formed sequence analysis of the UGT gene in strain UGTPg45 and UGTPn50. The results demonstrated that ZWDRH2-8 and identified two missense mutations −1 −1 the kcat/Km of UGTPn50 (0.127 mM s )wassignificantly (Q222H and A322V), the UGT mutant was renamed − − higher than that of UGTPg45 (0.074 mM 1 s 1)(Table1). UGTPg45-HV. Catalytic kinetic analysis revealed that, Subsequently, UGTPn50 was integrated into the multi-copy although the Km value of UGTPg45-HV toward PPD delta DNA site of ZW04BY-RS using the strong promoter and UDP-glucose remained almost unchanged, the Vmax UAS-TDH3, yielding strain ZWDRH2-7. The ginsenoside was nearly doubled and the kcat/Km value increased by Rh2 yield of ZWDRH2-7 was 145.7 mg/L in shake flasks, more than 40% compared with that of UGTPg45 twofold greater than that of ZWDRH2-3 (Fig. 4a). (Table 1). The significant improvement in catalytic activity may explain the increase of ginsenoside Rh2 Screening of UGTPg45 mutants derived from direct production in strain ZWDRH2-8 (Fig. 4b). evolution with high PPD catalysis efficiencies for In order to explore the molecular mechanism for the ginsenoside Rh2 production catalytic efficiency promotion of the mutant UGTPg45- Protein engineering by direct evolution is a powerful HV, the wild-type UGTPg45 was analyzed under strategy for engineering proteins without three- homology modeling and ligand docking. The analyzing

Fig. 4 UDP-glycosyltransferase (UGT) biopart optimization for improving ginsenoside Rh2 production. a Production of ginsenoside Rh2 and related compounds by novel UGT genes from different sources. UGTPn50, UGT biopart from P. notoginseng; UGT73F3, UGT biopart from M. truncatula; and UGT73C10, UGT biopart from B. vulgaris. The UGT biopart UGTPn50, with the highest ginsenoside Rh2 production, was integrated into the multi- copy delta DNA sites under the control of the UAS-TDH3 promoter, yielding strain ZWDRH2-10. b Production of ginsenoside Rh2 and related compounds by protein engineering of UGTPg45. The error bars indicate the SEMs of three biological replicates Wang et al. Cell Discovery (2019) 5:5 Page 7 of 14

Table 1 The kinetic parameters of engineered and new identiﬁed UGT bioparts toward PPD and UDP-glucose as substrates

a -1 -1 -1 Enzyme Vmax (nmol/(min mg)) Km (mM) kcat (s ) kcat/Km (mM s )

UGTPg45 8.69 ± 2.26 0.105 ± 1.06 × 10–2 0.0076 ± 2.03 × 10–3 0.074 ± 0.027 UGTPg45-HV 15.91 ± 1.07 0.137 ± 5.65 × 10–2 0.0139 ± 1.07 × 10–3 0.109 ± 0.037 UGTPn50 21.81 ± 2.67 0.152 ± 4.48 × 10–2 0.0189 ± 2.70 × 10–3 0.127 ± 0.020 UGTPn50-HV 31.71 ± 0.92 0.149 ± 4.04 × 10–2 0.0275 ± 1.37 × 10–3 0.190 ± 0.042 aAll given data in this table representing mean values from two or three repeats with corresponding standard deviations data demonstrated that the two mutated sites on E323 of UGTPg45 (Fig. 5a). Because the two amino acid UGTPg45-HV were not included in the binding pocket, mutations Q222H and A322V contributed to the in which the enzyme directly contact with the substrates increased activity of UGTPg45-HV, we are wondering PPD or UDP-glucose (Supplementary Fig. S4). Thus, whether these two amino acids mutations will also molecular dynamics (MD) simulations were further per- improve the performance of UGTPn50. Thus, the formed on UGTPg45 and UGTPg45-HV, respectively. A322V mutation and the other missing amino acid E in The root mean square deviation (RMSD) showed 60 ns UGTPn50 were inserted between E321 and A322 of was sufficient for both UDP-glucose and PPD to come UGTPn50, yielding the UGT mutant UGTPn50-VE. into stability (Supplementary Fig. S5a). Although the two Moreover, we constructed the single mutant UGTPn50- mutated sites of UGTPg45-HV were supposed to be Q222HandacombinedmutantUGTPn50-HV outside the substrate-binding pocket (Supplementary (UGTPn50-Q222H-VE). Then, these three mutants Fig. S4), the result of MD simulation did show that the were each integrated into the chromosome of ZW04BY- two mutated amino acids could significantly affect the RS. The resulting strains ZWDRH2-6A, ZWDRH2-6B, binding affinities between the enzyme and the substrate and ZWDRH2-6C produced 45.5, 55.6, and 80.0 mg/L of PPD. Calculated binding-free energy of PPD in ginsenoside Rh2, respectively (Fig. 5b).Ofthese,the UGTPg45-HV was −45.22 ± 3.45 kcal/mol, which was ginsenoside Rh2 production of strain ZWDRH2-6C was much lower than that in UGTPg45 (−33.29 ± 3.54 kcal/ 75.4% greater than that of strain ZWDRH2-6 with wild- mol). This means the two mutated amino acids make type UGTPn50 as the UGT biopart. Therefore, the UGTPg45-HV binding more tightly with substrate PPD Q222H mutation and two amino acids VE insertion than UGTpg45. Decompositions on the two total free significantly improved the performance of UGTPn50 energies (Supplementary Fig. S6) showed the substrate in vivo. Ginsenoside Rh2 production was much higher PPD was binding differently in the pocket, which could inZWDRH2-6Cthaninthestrainswiththetwo also be reflected in the averaged structures during the last mutant UGT bioparts (ZWDRH2-6A and ZWDRH2- 10 ns trajectories (Supplementary Fig. S5b–d). These 6B). The same trend was found when comparing gin- analyses indicate the two mutated amino acids of senoside Rh2 production of the two single mutations UGTPg45-HV outside the pocket may alter the config- of UGTPg45, UGTPg45-Q222H (strain ZWDRH2-1A) uration of its substrate binding pocket. The above data and UGTPg45-A322V (strain ZWDRH2-1B) with may explain the higher catalyzing capacity of the mutant UGTPg45-HV (strain ZWDRH2-8) (Fig. 5b). These enzyme UGTPg45-HV. results suggest that there is a synergistic effect between UGTPg45-HV was then integrated into the multi- these two amino acid mutations in improving ginseno- copy delta DNA sites under the control of the stronger side Rh2 production. UAS-TDH3 promoter to construct strain ZWDRH2-9. Combining these results, we acquired the improved UGT The ginsenoside Rh2 yield of ZWDRH2-9 increased to biopart UGTPn50-HV, and catalytic kinetic analysis 150.9 mg/L, more than threefold that of ZWDRH2-1 revealed that both the Vmax and kcat/Km of UGTPn50-HV (Fig. 4b). increased by more than 40% compared with UGTPn50 (Table 1). Therefore, the final Rh2 cell factory ZWDRH2-10 Combining the engineered UGTPg45-HV and novel biopart was constructed by integrating UGTPn50-HV into the delta UGTPn50 to increase ginsenoside Rh2 production DNA sites under the control of the UAS-TDH3 promoter. TheencodinggeneofUGTPn50fromP. notoginseng Rh2 production by ZWDRH2-10 reached 179.3 mg/L in sharedahighsequenceidentitywiththatofUGTPg45. shake flasks, which was fivefold that of strain ZWDRH2- Compared with UGTPg45, two amino acids were 1(Table2). To the best of our knowledge, this is the highest missing in UGTPn50, which corresponded to A322 and Rh2 production ever reported in a shake flask system. Wang et al. Cell Discovery (2019) 5:5 Page 8 of 14

Fig. 5 Combination of the engineered UGTPg45-HV and novel biopart UGTPn50 to further improve ginsenoside Rh2 production. a Amino acid sequence alignment of UGTPn50, UGTPg45, and UGTPg45-HV. b Production of ginsenoside Rh2 by different UGT bioparts in the PPD chassis strain. The error bars indicate the SEMs of three biological replicates

Fed-batch fermentation enabled gram-scale production of was 76.9 mg/g dcw (i.e., 7.7% of the dry weight of yeast), PPD and ginsenoside Rh2 much higher than the total saponin content of 5-year-old Fed-batch fermentation can promote high cell densities ginseng root (2– 3%)8. and has been proven to boost the production of many For fed-batch fermentation of the ginsenoside Rh2 cell valuable natural products, such as artemisinic acid, tan- factory strain ZWDRH2-10, the fermentation para- shinone, and ginsenosides, in engineered yeast1,18,31.We meters were controlled as the same for ZW04BY-RS. first performed fed-batch fermentation of the PPD- Fermentation was completed within 144 h, and the final producing chassis ZW04BY-RS in a 10 L bioreactor. The biomass was 157.3 g/L dcw. The final DM, PPD, and fermentation was completed within 144 h, and the max- ginsenoside Rh2 titers were 8.10, 9.05, and 2.25 g/L, imum biomass reached 143.0 g/L dry cell weight (dcw). respectively (Table 3). This is the highest production The final DM and PPD titers reached 7.1 g/L and 11.02 g/ and first report of gram-scale production of ginsenoside L, respectively (Table 3), representing the highest PPD Rh2,nearlysevenfoldthatofZY-7reportedbyZhuang production ever reported. The dry weight content of PPD et al17. Wang et al. Cell Discovery (2019) 5:5 Page 9 of 14

Table 2 Ginsenosides production by different strains in shake ﬂasks

Strainsa Rh2 (mg/L) PPD (mg/L) DM (mg/L)

ZW03BY – 237.9 ± 9.8 2.7 ± 0.5 ZW04BY – 329.7 ± 4.4 195.9 ± 7.7 ZW04BY-RS – 529.0 ± 25.5 72.6 ± 2.1 ZWDRH2–1 35.7 ± 3.8 523.2 ± 8.3 67.3 ± 15.7 ZWDRH2-2 52.7 ± 2.5 360.3 ± 13.2 122.3 ± 19.1 ZWDRH2-3 74.7 ± 5.1 375.4 ± 7.4 87.7 ± 9.5 ZWDRH2-4 0 563.4 ± 21.0 88.0 ± 7.4 ZWDRH2-5 14.1 ± 0.1 557.6 ± 48.9 56.7 ± 4.7 ZWDRH2-6 45.6 ± 0.5 478.6 ± 19.4 56.4 ± 4.1 ZWDRH2-7 145.7 ± 4.7 306.7 ± 0.4 184.5 ± 10.7 ZWDRH2-8 60.5 ± 2.5 514.9 ± 2.6 55.8 ± 1.5 ZWDRH2-9 150.9 ± 10.0 269.7 ± 14.6 190.9 ± 48.9 ZWDRH2-1A 38.4 ± 4.2 500.6 ± 5.9 77.3 ± 9.7 ZWDRH2-1B 38.7 ± 7.7 515.2 ± 7.2 65.5 ± 1.3 ZWDRH2-6A 45.6 ± 5.2 475.3 ± 11.0 90.3 ± 20.7 ZWDRH2-6B 55.6 ± 2.5 369.3 ± 1.9 109.0 ± 9.1 ZWDRH2-6C 80.0 ± 6.6 303.3 ± 16.3 174.3 ± 8.8 ZWDRH2-10 179.3 ± 6.5 255.3 ± 19.0 195.3 ± 21.8 aAll given data in this table representing mean values from three repeats with corresponding standard deviations

Table 3 Ginsenosides production by different strains in 10 L bioreactor

Strains Biomass (g/L DCW) Rh2 (mg/L) PPD (mg/L) DM (mg/L)

ZW04BY-RS 143.0 – 11017.2 7103.4 ZWDRH2-10 157.3 2252.3 9054.5 8088.8

Discussion the chassis cell for the efficient production of ginsenoside Due to the complexity of living systems, the developing Rh2, we aimed to systemically enhance the biosynthetic of a potent artificial biological system with desired func- flux of PPD via overexpressing all the 10 pathway genes. tion and efficiency is challenging32. In the field of natural We divided the pathway into two modules and adopted products production by synthetic biological approaches, the modular engineering strategies for PPD chassis strain many hurdles have to be overcome in order to acquire an construction34,35. A single-step integration of the efficient cell factory. Such as the developing of a robust downstream-module resulted in the production of more chassis strain for sufficient precursor supply, and the than 200 mg/L of PPD in shake flasks (Fig. 2), which characterizing and tuning the compatibilities among dif- demonstrated the feasibility of modular engineering. The ferent bioparts as well as the compatibilities between key subsequent integration of the upstream-module resulted bioparts and the chassis strain33. in more than double of total triterpenoid (PPD + DM) To build a robust chassis cell with high-level precursor production (Fig. 2). Finally, by increasing the expression supply, it is necessary to trap more metabolic flux to the level of the rate-limited enzyme P450 to convert DM into related target pathway, modulate the flux distribution PPD, the PPD titer reached to more than > 500 mg/L in between the native and heterogeneous pathway and shake flasks and > 10 g/L in fed-batch fermentation (Fig. 2 optimize the rate-limited enzymes. In this study, in order and Table 3), which is the highest PPD production and to construct a high-level PPD-producing yeast strain as about 2.5-fold of the previously reported highest one20. Wang et al. Cell Discovery (2019) 5:5 Page 10 of 14

Due to the long-term evolution, some natural bioparts specificity to the C3–OH of triterpenoids facing the isolated from native host may suffer from poor expression problems of activities or by-products production in level, low activities, and other undetectable problems in microbial cell factories. the microbial chassis cells, which is referred to as biopart- Protein engineering provides another approach to incompatibility32. For ginsenoside Rh2 cell factories, the improve the compatibility of bioparts with chassis cells. In incompatibility of the key UGT biopart UGTPg45 with this study, we adopt the in vivo directed evolution strategy the PPD-producing chassis (i.e., poor expression and low for the protein engineering of UGTPg45. Compared with glycosylation efficiency) extremely limited the final gin- previous commonly used in vitro directed evolution senoside Rh2 yield. Besides, the introducing of UGTPg45 methods, in which the screened and selected mutants into the PPD-producing yeast chassis dramatically were then transformed into the chassis, the superiority of reduced its total triterpenoids production14. To screen an in vivo directed evolution is obvious38,39. Because the alternative bioparts with higher compatibility from other mutants are introduced directly into the chassis and source has usually been applied in the optimization of cell screened according to the yield of a target product in vivo, factory21,36. In fact, some UGTs from bacteria (e.g., the selected mutated bioparts could be improved with not UGT109A1 and Bs-YjiC) and yeast (UGT51) have been only better enzymatic characteristics but also better per- – adopted to produce ginsenoside Rh215 17. The original formance in compatibility with chassis. In this study, a substrates of those UGTs might be quite different from UGT mutant UGTPg45-HV with two missense mutations PPD, however, owing to their lower substrate specificity (Q222H and A322V), which gave 70% increase of ginse- and even lower regio-specificity, those UGTs turn to be noside Rh2 yield compared with UGTPg45, was acquired potential bioparts for ginsenoside Rh2 synthesis. Never- via in vivo directed evolution (Fig. 4b). The catalytic theless, the substrate flexibility or low regio-specificity of efficiency (kcat/Km) of UGTPg45-HV increased by more those UGTs from microbes often resulted in the pro- than 40%, which may explain the increase in ginsenoside duction of various by-products16,17. Some plant UGTs Rh2 production (Table 1). exhibit lower substrate specificity but high regio-specifi- MD simulation analyses demonstrated the two mutant city, i.e., these UGTs might catalyze the glycosylation of amino acids outside the binding pocket might affect the different aglycons with similar structure at the same performance of UGTPg45 by altering the configuration position. For example, four UGTs, UGT73C10, of its substrate binding pocket. As shown in the illus- UGT73C11, UGT73C12 and UGT73C13, from B. vulgaris tration of the averaged structures (Supplementary could glucosylate the C3–OH of different triterpenoids Fig. S5b–d), PPD was binding deeper in the binding including oleanolic acid, hederagenin, and betulinic pocket of UGTPg45-HV, which meant the pocket was acid30. Those plant UGTs catalyzing the C3–OH glyco- less open in the mutant enzyme. In addition to the sylation of triterpenoids with high regio-specificity pro- stronger binding between UGTPg45-HV and PPD, PPD vide potential alternative bioparts for ginsenoside might get a more favorable pose in the narrower pocket Rh2 synthesis. Besides, P. notoginseng, another ginseno- of the mutant and thus leaded to the higher catalytic side producing plant, may also provide potential UGT efficiency of UGTPg45-HV. After the two beneficial bioparts for ginsenoside Rh2 synthesis37. In this work, mutated sites were introduced into UGTPn50, the three UGTs, UGT73F3 from M. truncatula29, UGT73C10 optimized biopart UGTPn50-HV also demonstrated from B. vulgaris30, and UGTPn50 from P. notoginseng better in vivo performance and gave the highest ginse- were characterized to catalyze the C3–OH glycosylation noside Rh2 production in the same PPD-producing of PPD yielding ginsenoside Rh2 for the first time chassis (Fig. 5). (Fig. 4b). UGTPn50 and UGTPg45 belong to UGT74- In conclusion, we constructed the PPD-producing subfamily, while UGT73F3 and UGT73C10 belong to chassis strain ZW04BY-RS, which yielded a PPD titer of UGT73 subfamily. The sequence similarity of UGT73F3 529.0 mg/L in a shake flask system and 11.02 g/L in a 10 L and UGT73C10 with UGTPg45 are both less than 30%. bioreactor (Fig. 2 and Table 3). Based on this strain, we The results indicated that UGTs with low sequence developed a ginsenoside Rh2 cell factory with a ginseno- similarity might share similar functions. When introdu- side Rh2 titer of 179.3 mg/L in a shake flask system and cing these three UGTs into the PPD chassis, UGT73F3 2.25 g/L in a 10 L bioreactor (Fig. 5 and Table 3). To the with the lowest bioactivity resulted in undetected ginse- best of our knowledge, this is not only the highest gin- noside Rh2 production, UGTPn50 gave the highest ginsenoside Rh2 yield in engineered microbes, but also the senoside Rh2 yield, more than 25% increase compared highest yield of natural products with glycosylation with that of UGTPg45. UGT73C10, resulted in a medium modification. We believe the gram-scale per liter yield of ginsenoside Rh2 yield (14.1 mg/L) (Fig. 4a) and the pro- ginsenoside Rh2 by the fermentation of ZWDRH2-10 duction of a by-product 3-DMG (9.3 mg/L). It seems even could lower the Rh2 manufacturing cost greatly compared a plant UGT with low substrate specificity but high regio- with the traditional ginseng extraction and Wang et al. Cell Discovery (2019) 5:5 Page 11 of 14

biotransformation methods. However, it is necessary to Enzymatic assays of UDP-glycosyltransferases carry out pilot plant tests in larger fermenters before Enzymatic assays for glycosyltransferases were per- commercial production. formed as described previously14,40. Generally, the reaction was carried out in a 100 μL system containing 100 Materials and methods mM phosphate buffer (pH 7.5), 1% Tween-20, 5 mM Materials, plasmids, and strains UDP-glucose, 0.5 mM substrate, and 50 μL of UGT crude Authentic ginsenoside samples compound K, ginse- enzyme (~ 400 ng/mL) for 2 h in a 35 °C water bath. noside Rh2, F2, Rg3, Rd, Rb1, Rb2, and protopanaxadiol Reaction products was extracted by adding 100 μLofn- (PPD) were purchased from Nanjing Zelang Medical butanol, the organic phase was then evaporated and dis- Technology Co., Ltd. (Nanjing, China). Dammarenediol solved in methanol for TLC and HPLC analysis. II (DM) was purchased from BioBioPha Co., Ltd. For the kinetic study of UDP-glycosyltransferases, the (Kunming, China). Plant materials P. notoginseng and M. His-tagged UGTs in the crude enzyme were firstly quan- truncatula were kindly provided by Prof. Weiming Cai tified by dot bolt as described previously41. The purified C- and Prof. Fang Xie, respectively. Both of them are from terminal 6×His-tagged UDP-glycosyltransferase OleD the Institute of Plant Physiology and Ecology, SIBS, (Genbank accession no. ABA42119.2) was diluted serially CAS. E. coli strain TOP10 was used for gene cloning, (8, 12, 14, 16, 18, 32 ng/μL) to make a standard curve for and BL21 (DE3) was used for UGTs heterologous protein quantitative. The reaction mixtures contained 100 expression. S. cerevisiae strain BY4742 (MATα,his3Δ1, mM phosphate buffer (pH 7.5), 1% Tween-20, 0.5 mM leu2Δ0, lys2Δ0, ura3Δ0)obtainedfromEUROSCARF PPD, 40-300 μM UDP-glucose, and 60 µl crude enzyme of was used as the parent strain for all engineering. Codon- UGTs (~ 400 ng/mL) in a final volume of 300 μL. The optimized genes synPgDDS, synPgPPDS, and synPgCPR1 reactions were incubated at 30 °C for 30 min. HPLC ana- (Genbank accession nos. ACZ71036.1, AEY75213.1, and lysis was used to quantify the target product in each AIC73829.1, respectively) were synthesized by Gen- reaction. The Michaelis–Menten parameters were calcu- script Corporation (Nanjing, China). All the strains used lated by Lineweaver–Burk plot. All data are presented as or constructed in this study are listed in Supplementary means ± SD of two or three independent repeats. Table S1 and the primers used for the construction of the plasmids and strains are listed in Supplementary Construction of yeast strains Table S2. The general procedure for each yeast strain construction was descript as following. First, each promoter, gene, Cloning, synthesis, and heterologous expression of UDP- terminator, selection marker, and homologous arm was glycosyltransferases PCR amplified to give the basic fragments, the adjacent The coding sequence of UDP-glycosyltransferase basic fragments all sharing 40–75 bp homologous UGT73F3 (Genbank accession no. ACT34898.1) was a sequences for recombination or fusion PCR; Second, PCR-amplified form of M. truncatula using primer using the adjacent 2–4 basic fragments as template, fusion 73F3-F and 73F3-R and cloned into the pMD18T vector PCR was performed to give each fusion fragments. Third, (Takara, Dalian, China). The coding sequence of the fusion fragments were purified, quantified, and co- UGTPn50 was a PCR-amplified form of P. notoginseng transformed into yeast strain via standard LiAc/ssDNA using primer Pn50-F and Pn50-R and cloned into the method, because each adjacent fragment sharing 40–75 pMD18T vector. UGT73C10 (Genbank accession no. bp homologous sequences they will be join together via AFN26666.1) was synthesized by Genscript yeast homologous recombination and integrated into Corporation. chromosome42. Heterologous expression of the UGT genes in E. coli was carried out as described previously14. Briefly, UGT Yeast cultivation and metabolites extraction genes with a C-terminally 6×His-tag was ligated into the Yeast strains were grown in YPD medium composed of pET-28a vector via ClonExpress II One Step Cloning Kit 10 g/L Bacto Yeast Extract (BD Difco, USA), 20 g/L Bacto (Vazyme Biotech Co., Ltd, Nanjing, China) and trans- peptone (BD Difco, USA), and 20 g/L dextrose (Sino- formed into E. coli BL21 (DE3). Protein expression was pharm Chemical, Shanghai, China). For auxotrophic conducted by 0.2 mM IPTG induction for 18 h at 18 °C. marker selection yeast clones were grown in SD medium The cells were collected by centrifugation and suspended composed of 6.7 g/L Difco Yeast Nitrogen Base without in 100 mM phosphate buffer (pH 7.5) supplemented with amino acids (BD Difco, USA), 20 g/L dextrose and 2 g/L 1 mM PMSF and then disrupted with a French Press (25 Drop-out Mix Synthetic minus appropriate amino acids kpsi). The cell debris was removed by centrifugation (Sigma-Aldrich, USA). For KanMx marker selection yeast (12,000 g, 20 min), and the supernatant was used for clones were grown in YPD medium containing 200 mg/L enzymatic assays. G418 sulfate (Sangon Biotech, Shanghai, China). Wang et al. Cell Discovery (2019) 5:5 Page 12 of 14

Shake flask fermentation of engineered yeast strains. equipped with a LC20ADXR pumper, an auto-sampler Individual clones were inoculated into the YPD medium and a diode array detector. Chromatographic separation and cultivated at 30 °C, 250 rpm for 16 h. Aliquots were of ginsenoside Rh2, PPD, and DM was carried out at 35 °C diluted to an initial OD600 of 0.05 in 10 mL of YPD on a Shim-pack XR-ODS column (100 mm × 2.0 mm, 2.2 medium in 50 mL shake flasks and grown at 30 °C, 250 μm, Shimadzu, Kyoto, Japan). The gradient elution system rpm for 96 h. The fermentation broth was extracted with consisted of water (A) and acetonitrile (B). Separation was n-butanol and used for the analysis of DM, PPD, and achieved using the following gradient: 0–2.5 min (60% B), ginsenoside Rh2. 2.5–10 min (60%–90% B), 10–11 min (90% B), and 11–13 Fed-batch fermentation of yeast strain ZW04BY-RS and min (60% B), and the flow rate was kept at 0.45 mL/min. ZWDRH2-10 were conducted in a 10 L bioreactor (T & J For rapid detection products of mutants of UGTPg45, Bioengineering, Shanghai, China). The medium for seed, Accucore C18 columns (2.6 μm, 2.1 mm × 100 mm, batch fermentation, and the control of feed rate was Thermo Scientific, USA) were used and separation was conducted as previously described1,18. About 0.3 L seed of achieved using the following gradient: 0–1 min (55% B), strains ZW04BY-RS or ZWDRH2-10 was inoculated into 1–3.5 min (90% B), and 3.5–5 min (55% B), and the flow 2.7 L of batch medium in the 10-L bioreactor and then rate was kept at 0.45 mL/min. Therefore, each sample can started the fermentation. Temperature of the fermentation be finished within 5 min, and a 96-well plate samples can was set at 30 °C. pH was controlled at 5.0 by addition of be finished within 8 h. ammonia hydroxide. Dissolved O2 was controlled above 30% by varying speed of agitation or air flow rate. Feeding Molecular dynamics simulations and following analyses rate was controlled by maintaining ethanol < 0.5 g/L. Initial structure of WT UGTPg45 was modeled with I- – Tasser web server43 45. PyMOL v1.746 was used to gen- In vivo direct evolution of UGTPg45 erate the mutant UGTPg45-HV. UDP-glucose was placed Random mutagenesis of UGTPg45 (Genbank accession in the binding pocket in the reference of a previous no. AKA44586.1) by error-prone PCR was performed released crystal structure of the complex of a UGT using the GeneMorph II random mutagenesis kit (Agilent enzyme and UDP-glucose (PDB code: 2ACW47). PPD was Technologies Inc., CA, USA). The PCR reaction condition docked and refined into the binding pocket with Auto- was set as suggested in the user manual, error rate was Dock Vina v1.1.248. MD simulations and most analysis controlled to be 1–2 mutations per gene. The UGTPg45 procedures were conducted using the Amber18 software mutants was transformed into PPD-producing yeast package49. Hydrogen atoms were added using the LEaP ZW04BY-RS to generate a mutant library (detailed pro- module. Counter-ions were used to maintain system cedures to construct strain library are provided in neutrality. All systems were solvated in a truncated the Supplementary Information). The clones that were octahedron box of TIP3P waters with a buffer of 10 Å. picked from plates are incubated into deep 96-well plates, The ff14SB force field50 was used for proteins. The para- which contained 600 μL of YPD medium and cultivated at meters of ligands were created using Antechamber with 30 °C, 280 rpm for 96 h. At the end of fermentation, 600 GAFF as basic force field. All the MD simulations were μL n-butanol was added into each well, and PPD and Rh2 accelerated with the CUDA version of PMEMD and run ® ® were extracted. The organic phase was transformed into a on NVIDIA Tesla P100. Up to 20000-step, steepest new 96-well plate for quantification of PPD and Rh2 by descent minimization was performed to relieve any fur- HPLC. ther structural clash in the solvated systems. This was followed by a 400-ps heating up and a 200-ps equilibra- Chemical analysis tion in the NVT ensemble at 298K. The 60 ns’ production TLC analysis. The TLC analysis was performed using runs were simulated in the NPT ensemble at 298K with silica gel 60 F254 plates (Merck KGaA, Darmstadt, Ger- a time step of 2 fs in the Berendsen’s thermostat many) with chloroform/methanol/water (30/10/1, v/v) as and barostat with default settings. Detailed simulation the developing solvent. The spots on the TLC plates were information was listed in Supplementary Table S3. visualized by spraying with 1% (w/v) vanillin in H2SO4/ Binding-free energies were calculated with MMPBSA with ethanol (6/100, v/v) followed by heating at 110 °C for 5 sampling from the last 10 ns of each trajectory, which got min. For identification, a mixture of ginsenoside authentic into equilibrium in the production run. samples containing compound K, ginsenoside Rh2, F2, Rg3, Rb1, Rb2, and Rd, and PPD of concentration 0.2 mg/ Acknowledgements mL (dissolved in methanol) was also spotted on each This work was financially supported by the Natural Science Foundation of plate. China (No. 21672228), National Basic Research Program of China (2015CB755703), the Key Deployment Projects of the Chinese Academy of HPLC analysis. The HPLC analysis was performed on a Sciences (No. KFZD-SW-215), the Strategic Biological Resources Service Shimadzu LC20A system (Shimadzu, Kyoto, Japan) Network Plan of the Chinese Academy of Sciences (ZSYS-016), the Wang et al. Cell Discovery (2019) 5:5 Page 13 of 14

International Great Science Program of the Chinese Academy of Sciences (No. 14. Wang, P. P. et al. Production of bioactive ginsenosides Rh2 and Rg3 by 153D31KYSB20170121), the Natural Science Foundation of Shanghai metabolically engineered yeasts. Metab. Eng. 29,97–105 (2015). (16ZR1441900), the National Mega-project for Innovative Drugs 15. Liang, H. et al. Production of a bioactive unnatural ginsenoside by metabo- (2017ZX09101003-006-002) and the Strategic Priority Research Program of the lically engineered yeasts based on a new UDP-glycosyltransferase from Bacillus Chinese Academy of Sciences (XDB27020206). subtilis. Metab. Eng. 44,60–69 (2017). 16. Dai, L. et al. One-pot synthesis of ginsenoside Rh2 and bioactive unnatural Author details ginsenoside by coupling promiscuous glycosyltransferase from Bacillus subtilis 1CAS-Key Laboratory of Synthetic Biology, CAS Center for Excellence in 168 to sucrose synthase. J. Agric. Food Chem. 66, 2830–2837 (2018). Molecular Plant Sciences, Institute of Plant Physiology and Ecology, Shanghai 17. Zhuang, Y. et al. Biosynthesis of plant-derived ginsenoside Rh2 in yeast via Institutes for Biological Sciences, Chinese Academy of Sciences, 300 Fenglin Rd, repurposing a key promiscuous microbial enzyme. Metab. Eng. 42,25–32 Shanghai 200032, China. 2University of Chinese Academy of Sciences, Beijing (2017). 100049, China. 3Bio-Med Big Data Center, CAS Key Laboratory of 18. Dai, Z. et al. Metabolic engineering of Saccharomyces cerevisiae for production Computational Biology, CAS-MPG Partner Institute for Computational Biology, of ginsenosides. Metab. Eng. 20, 146–156 (2013). Shanghai Institute of Nutrition and Health, Shanghai 200031, China 19. Zhao, F. L. et al. Optimization of a cytochrome P450 oxidation system for enhancing protopanaxadiol production in Saccharomyces cerevisiae. Bio- – Author contributions technol. Bioeng. 113, 1787 1795 (2016). P.P.W., Z.H.Z., and X.Y. conceived and designed the project. P.P.W., W.W., W.Y., X. 20. Zhao, F. L. et al. Enhancing Saccharomyces cerevisiae reactive oxygen species D.L., W.F.Z., C.S.Y., and C.J.L. designed and performed all the experiments and and ethanol stress tolerance for high-level production of protopanoxadiol. – analyzed the data. P.P.W., Z.H.Z., X.Y., and W.Y. wrote the paper. All authors Bioresour. Technol. 227,308 316 (2017). contributed to the discussion and approved the final paper. 21. Dai,Z.etal.Producingaglycons of ginsenosides in bakers' yeast. Sci. Rep. 4, 3698 (2014). 22. Jensen, N. B. et al. EasyClone: method for iterative chromosomal integration of Conflict of interest multiple genes in Saccharomyces cerevisiae. Fems.Yeast.Res.14, 238–248 (2014). The authors declare that they have no conflict of interest. 23. Stovicek, V., Borja, G. M., Forster, J. & Borodina, I. EasyClone 2.0: expanded toolkit of integrative vectors for stable gene expression in industrial Sacchar- omyces cerevisiae strains. J. Ind. Microbiol. Biot. 42,1519–1531 (2015). Publisher’s note 24. Blazeck, J., Garg, R., Reed, B. & Alper, H. S. Controlling promoter strength and Springer Nature remains neutral with regard to jurisdictional claims in regulation in Saccharomyces cerevisiae using synthetic hybrid promoters. Bio- published maps and institutional affiliations. technol. Bioeng. 109, 2884–2895 (2012). 25. Bowles,D.,Isayenkova,J.,Lim,E.K.&Poppenberger,B.Glycosyltransferases: Supplementary Information accompanies the paper at (https://doi.org/ managers of small molecules. Curr. Opin. Plant. Biol. 8, 254–263 (2005). 10.1038/s41421-018-0075-5). 26. Vogt, T. & Jones, P. Glycosyltransferases in plant natural product synthesis: characterization of a supergene family. Trends Plant. Sci. 5,380–386 (2000). Received: 27 July 2018 Revised: 19 October 2018 Accepted: 2 November 27. Moses, T. et al. Combinatorial biosynthesis of sapogenins and saponins in 2018 Saccharomyces cerevisiae using a C-16 alpha hydroxylase from Bupleurum falcatum. Proc.Natl.Acad.Sci.U.S.A.111,1634–1639 (2014). 28. Fu, X. et al. Antibiotic optimization via in vitro glycorandomization. Nat. Bio- technol. 21, 1467–1469 (2003). References 29. Naoumkina,M.A.etal.Genomicandcoexpressionanalysespredictmultiple 1. Paddon, C. J. et al. High-level semi-synthetic production of the potent anti- genes involved in triterpene saponin biosynthesis in Medicago truncatula. malarial artemisinin. Nature 496, 528–532 (2013). Plant Cell 22,850–866 (2010). 2. Ro, D. K. et al. Production of the antimalarial drug precursor artemisinic acid in 30. Augustin, J. M. et al. UDP-glycosyltransferases from the UGT73C subfamily in engineered yeast. Nature 440,940–943 (2006). Barbarea vulgaris catalyze sapogenin 3-O-glucosylation in saponin-mediated 3. Luo, Y. et al. Engineered biosynthesis of natural products in heterologous insect resistance. Plant Physiol. 160, 1881–1895 (2012). hosts. Chem. Soc. Rev. 44,5265–5290 (2015). 31. Guo, J. et al. CYP76AH1 catalyzes turnover of miltiradiene in tanshinones 4. Sun, X. X. et al. Synthesis of chemicals by metabolic engineering of microbes. biosynthesis and enables heterologous production of ferruginol in yeasts. Proc. Chem. Soc. Rev. 44, 3760–3785 (2015). Natl. Acad. Sci. U. S. A. 110, 12108–12113 (2013). 5. Park,J.A.,Lee,K.Y.,Oh,Y.J.,Kim,K.W.&Lee,S.K.Activationofcaspase-3 32. Kwok, R. Five hard truths for synthetic biology. Nature 463,288–290 (2010). protease via a Bcl-2-insensitive pathway during the process of ginsenoside 33. Mitchell, W. Natural products from synthetic biology. Curr.Opin.Chem.Biol.15, Rh2-induced apoptosis. Cancer Lett. 121,73–81 (1997). 505–515 (2011). 6. Kim, H. S. et al. Effects of ginsenosides Rg3 and Rh2 on the proliferation of 34. Zhou, Y. J. J. et al. Modular pathway engineering of diterpenoid synthases and prostate cancer cells. Arch. Pharm. Res. 27,429–435 (2004). the mevalonic acid pathway for miltiradiene production. J. Am. Chem. Soc. 7. Wong, A. S. T., Che, C. M. & Leung, K. W. Recent advances in ginseng as cancer 134, 3234–3241 (2012). therapeutics: a functional and mechanistic overview. Nat. Prod. Rep. 32, 35. Biggs,B.W.,DePaepe,B.,Santos,C.N.S.,DeMey,M.&Ajikumar,P.K. 256–272 (2015). Multivariate modular metabolic engineering for pathway and strain optimi- 8. Shibata, S. Chemistry and cancer preventing activities of ginseng saponins zation. Curr. Opin. Biotech. 29,156–162 (2014). and some related triterpenoid compounds. J. Korean Med. Sci. 16,S28–S37 36. Wang,J.F.,Li,S.Y.,Xiong,Z.Q.&Wang,Y.Pathwaymining-basedintegration (2001). of critical enzyme parts for de novo biosynthesis of steviolglycosides sweet- 9. Han, B. H. et al. Degradation of ginseng saponins under mild acidic conditions. ener in Escherichia coli. Cell Res. 26,258–261 (2016). Planta Med. 44,146–149 (1982). 37. Kim,D.H.ChemicaldiversityofPanax ginseng, Panax quinquifolium, and Panax 10. Bae, E. A., Han, M. J., Kim, E. J. & Kim, D. H. Transformation of ginseng saponins notoginseng. J. Ginseng Res. 36,1–15 (2012). to ginsenoside Rh2 by acids and human intestinal bacteria and biological 38. Xie,W.P.,Lv,X.M.,Ye,L.D.,Zhou,P.P.&Yu,H.W.Constructionoflycopene- activities of their transformants. Arch. Pharm. Res. 27,61–67 (2004). overproducing Saccharomyces cerevisiae by combining directed evolution and 11. Su, J.-H., Xu, J.-H., Lu, W.-Y. & Lin, G.-Q. Enzymatic transformation of ginsenoside metabolic engineering. Meta. Eng. 30,69–78 (2015). Rg3 to Rh2 using newly isolated. Fusarium proliferatum ECU2042. J. Mol. Catal. 39. Lv, X. M. et al. Combinatorial pathway optimization in Escherichia coli by B-Enzym. 38, 113–118 (2006). directed co-evolution of rate-limitingenzymesandmodularpathwayengi- 12. Dong, L., Xu, J., Feng, G., Li, X. & Chen, S. Soil bacterial and fungal community neering. Biotechnol. Bioeng. 113,2661–2669 (2016). dynamics in relation to Panax notoginseng deathrateinacontinuouscrop- 40. Yan, X. et al. Production of bioactive ginsenoside compound K in metaboli- ping system. Sci. Rep. 6, 31802 (2016). cally engineered yeast. Cell Res. 24,770–773 (2014). 13. Jung, S. C. et al. Two ginseng UDP-glycosyltransferases synthesize ginsenoside 41. Faoro,V.&Stanta,G.inGuidelines for Molecular Analysis in Archive Tissues Rg3 and Rd. Plant Cell Physiol. 55, 2177–2188 (2014). (Giorgio S., ed) 275–276 (Springer: Berlin, Heidelberg, 2011). Wang et al. Cell Discovery (2019) 5:5 Page 14 of 14

42. Shao, Z., Zhao, H. & Zhao, H. DNA assembler, an in vivo genetic method 47. Shao, H. et al. Crystal structures of a multifunctional triterpene/flavonoid gly- for rapid construction of biochemical pathways. Nucleic Acids Res. 37,e16 cosyltransferase from Medicago truncatula. Plant Cell 17,3141–3154 (2005). (2009). 48. Trott, O. & Olson, A. J. AutoDock Vina: improving the speed and accuracy of 43. Zhang, Y. I-TASSER server for protein 3D structure prediction. BMC Bioinf. 9,40 docking with a new scoring function, efficient optimization, and multi- (2008). threading. J. Comput. Chem. 31, 455–461 (2010). 44. Roy, A., Kucukural, A. & Zhang, Y. I-TASSER: a unified platform for automated 49. Lee,T.S.etal.GPU-acceleratedmoleculardynamicsandfreeenergymethods protein structure and function prediction. Nat. Protoc. 5,725(2010). in Amber18: performance enhancements and new features. J. Chem. Inf. 45. Yang, J. et al. The I-TASSER Suite: protein structure and function prediction. Model. https://doi.org/10.1021/acs.jcim.8b00462 (2018). Nat. Methods 12, 7 (2015). 50. Maier, J. A. et al. ff14SB: improving the accuracy of protein side chain and 46. DeLano, W. L. Pymol: An open-source molecular graphics tool. CCP4 Newsletter backbone parameters from ff99SB. J. Chem. Theory Comput. 11,3696–3713 On Protein Crystallography 40,82–92 (2002). (2015).